AI's aggregate effect on jobs so far has been modest — no mass unemployment — but its gains are uneven: productivity rises concentrate with skilled workers while routine and low-skill roles face displacement, heightening labor-market polarization and calls for targeted upskilling and policy support.
The role of generative artificial intelligence on labor market: literature review
semantic_scholar
review_meta
medium evidence
7/10 relevance
Full text
usable extracted full text
DOI
Source
PDF
A systematic review of recent empirical studies finds AI's net impact on employment to date is modest and context-dependent: AI tends to complement skilled workers and raise productivity while displacing routine and low-skill roles, contributing to labor-market polarization.
The purpose of the study . This systematic literature review examines the role of AI in the labor market and its effectiveness in terms of productivity and employment outcomes. Methodology. We reviewed recent studies from 2020 to 2025 across global and regional contexts to assess how AI adoption influences job creation, displacement, and workforce composition. The objective was to synthesize current evidence on whether AI augments human labor or automates it away, and under what conditions. Using a systematic methodology, we analyzed 17 key publications from peer-reviewed academic journals. Originality / value. Our review finds that AI’s net impact on employment has so far been modest, with no clear evidence of mass unemployment caused by AI. However, AI-driven automation has uneven effects: it displaces certain routine and low-skill jobs while creating new possibilities for high-skill tasks, thus contributing to labor market polarization. Notably, AI tends to complement and enhance the productivity of skilled employees, whereas low-skilled roles face significant automation risk. Findings. In our discussion, we highlight the following findings: agreement that AI demands workforce upskilling and policy support, alongside divergent results, for example, conflicting evidence on net job creation in different contexts. A meta-analysis of the literature reveals surging research interest in 2023–2025 and a focus largely on advanced economies. Finally, we discuss implications: while AI can enhance labor productivity and create value, proactive measures are needed to ensure these gains translate into broad-based employment benefits. The review identifies research gaps such as limited studies in low-income countries and long-term generative AI effects and underscores the importance of policies to manage AI’s workforce transition.
Summary
Main Finding
The systematic review (17 studies, 2020–Apr 2025) finds that generative AI and related AI technologies have so far produced modest net employment effects at the aggregate level, but strongly heterogeneous impacts across occupations, sectors, skill groups and regions. AI is predominantly skill‑biased: it complements and raises productivity for high‑skill workers while substituting routine and low‑skill tasks, contributing to labor‑market polarization. Short‑to‑medium‑run displacement and wage pressures can occur locally even where long‑run employment and GDP grow.
Key Points
- Net employment: No clear mass unemployment to date; evidence points to reallocation rather than wholesale job loss. Aggregate effects are small but distributional effects large.
- Skill bias and polarization: AI increases demand for medium/high‑skill tasks and reduces routine/low‑skill roles, intensifying occupational polarization.
- Sectoral heterogeneity:
- Manufacturing: several firm‑level studies (esp. China) show AI/digital adoption can be labor‑augmenting via scale/output growth — employment can rise even as low‑skill tasks are automated.
- Services & white‑collar: generative AI and LLM exposure reach many high‑skill, cognitive occupations, creating automation risks for some non‑routine tasks.
- Geography: Research concentration in advanced economies; limited evidence for low‑income countries. Developing countries risk constrained "good job" creation, affecting development trajectories.
- Temporal dynamics: Short/medium run can feature transitional unemployment and wage suppression; long‑run models often predict net job creation and higher output if complementary demand and human capital adjust.
- Distributional & social effects: Uneven gains risk socio‑political polarization; older and less‑educated workers are more adversely affected.
- Complementarity with skills & platforms: Workers and freelancers with AI skills command wage premiums; platform data show higher demand for AI‑related tasks.
- Side effects: Corporate responses to AI investment (higher tech/high‑skill costs) can influence tax behavior (evidence of increased tax avoidance in some samples).
- Research trend: Surge in publications 2023–2025, driven by interest in generative AI tools (e.g., GPT‑based exposure studies).
Data & Methods
- Review type: Systematic literature review (SLR) covering Jan 2020–Apr 2025. Search: Scopus, Web of Science, institutional reports (ILO, WEF, OECD, World Bank). Final in‑depth sample: 17 studies selected from ~70 candidates.
- Inclusion criteria: Explicit labor‑market focus on AI (including ML, robotics, generative AI), 2020–2025, global/regional scope, peer‑reviewed papers and reputable institutional reports.
- Extraction & synthesis: Thematic grouping (overall employment effects, skill bias, sectoral outcomes, policy implications) and metadata analysis (year, region, methodology).
- Types of evidence in included studies:
- Econometric panel analyses (firm‑level and regional/commuting‑zone level; e.g., Acemoglu & Restrepo style robot exposure).
- Occupation‑task exposure mapping using O*NET and modelled LLM capabilities (e.g., GPT‑exposure studies).
- Platform data analyses of freelance markets.
- Case studies and proof‑of‑concept deployments (NLP/automation in business workflows).
- Macro and CGE simulation models for long‑run effects.
- Delphi expert elicitation and scenario analyses.
- Limitations of evidence noted: heterogeneity in exposure measures (robot density, patent/task overlap, text‑based AI adoption measures), short follow‑up for newest generative AI impacts, geographic bias toward advanced economies, and relatively few longitudinal worker‑level datasets covering the generative AI era.
Implications for AI Economics
- For research:
- Prioritize causal micro‑evidence on worker outcomes (wages, hours, transitions) using linked employer–employee and administrative data to capture heterogeneity and dynamics.
- Expand empirical coverage to low‑ and middle‑income countries to assess development and offshoring implications of generative AI.
- Deepen task‑level and occupation‑level measurement of generative AI exposure (beyond patent/robot proxies), and track actual adoption vs. theoretical exposure.
- Study firm investment responses, complementarities between AI and human capital, and differential impacts across firm size and sector.
- Investigate non‑employment outcomes: job quality, worker well‑being, political attitudes and tax behavior linked to AI adoption.
- For policy and applied economics:
- Active labor‑market measures (re‑skilling/upskilling, lifelong learning, targeted retraining) are critical to translate productivity gains into broad employment benefits.
- Social protection and transition support are needed to mitigate short‑run displacement, especially for older and low‑educated workers.
- Education systems should emphasize complementary skills (complex problem solving, digital literacy, domain expertise) and adaptability.
- Industrial and regional policies should consider where AI will be labor‑augmenting vs. substituting, supporting sectors and regions at risk of lost "good jobs."
- Tax and regulatory policy should account for firm behavior around AI investments (e.g., incentives, profit shifting, tax avoidance).
- For economic modeling and forecasting:
- Models must incorporate heterogeneous task complementarities, dynamic reallocation, and demand feedbacks (scale effects) rather than assuming uniform substitution.
- Scenario analysis should include distributional outcomes and transition frictions, not only long‑run equilibrium gains.
Overall, the literature to date suggests generative AI is reshaping the composition of work more than eliminating work at aggregate scale, but the distributional and regional effects are large and policy‑relevant. Future AI‑economics research should focus on causal, geographically broader, and worker‑level evidence to guide policies that make AI gains inclusive.
Assessment
Paper Typereview_meta
Evidence Strengthmedium — Synthesis of 17 recent peer‑reviewed empirical studies provides a useful overview, but the small sample, heterogeneous methods, concentration in advanced economies, and limited long-run/generative-AI evidence constrain the strength of aggregated conclusions about causality and net employment effects.
Methods Rigormedium — The work uses a systematic literature review and comparative meta-synthesis over a defined 2020–2025 window, but selection appears limited (17 studies), inclusion/exclusion criteria and search strategy are not fully detailed here, and heterogeneity across studies prevented a quantitative meta-analysis, reducing methodological rigor.
Sample17 peer‑reviewed publications published 2020–2025, selected for relevance to AI's labor-market and productivity impacts; studies cover global and regional contexts but are concentrated in advanced economies, with a surge of research in 2023–2025; underlying methods across the sample are heterogeneous (observational, quasi-experimental, case studies, firm- and worker-level analyses in some papers).
Themeslabor_markets productivity skills_training inequality adoption
GeneralizabilitySmall sample of studies limits representativeness, Geographic concentration in advanced economies; sparse evidence from low- and middle-income countries, Short-term focus with limited long-run or structural analyses, especially for generative AI, Heterogeneous study designs and outcomes hamper aggregation of causal effects, Sectoral coverage uneven — some industries under-studied
Claims (14)
| Claim | Direction | Outcome | Confidence & Evidence | Details |
|---|---|---|---|---|
| AI’s net impact on employment to date is modest — no clear evidence of mass unemployment. Employment | null_result | aggregate employment / unemployment rates |
Reading fidelity
medium
Study strength
medium
|
n=17
Net employment impact to date modest; no clear evidence of mass unemployment
|
| AI often complements and raises productivity for skilled workers and high-skill tasks. Firm Productivity | positive | productivity of skilled workers (e.g., output per worker, task-level productivity) |
Reading fidelity
medium
Study strength
medium
|
n=17
AI complements and raises productivity for skilled workers (reported in several studies)
|
| AI substitutes for and displaces many routine and low-skill occupations, increasing automation risk for those roles. Job Displacement | negative | employment levels in routine and low-skill occupations |
Reading fidelity
medium
Study strength
medium
|
n=17
AI substitutes for and displaces many routine and low-skill occupations (documented in multiple studies)
|
| AI contributes to labor‑market polarization: growth in high‑skill opportunities alongside contraction in many middle- and low‑skill roles. Inequality | mixed | occupational composition / wage distribution (polarization indicators) |
Reading fidelity
medium
Study strength
medium
|
n=17
Labor-market polarization: growth at high-skill end and contraction in middle/low-skill roles
|
| Whether AI is net job‑creating depends on context (sector, country, policy environment, and workforce skill composition). Employment | mixed | net employment effect (jobs created minus jobs displaced) by context |
Reading fidelity
medium
Study strength
medium
|
n=17
Net job effects depend on sector, country, policy, and workforce skill composition
|
| There is no consensus in the literature on net job effects — studies diverge on whether AI produces net job gains. Employment | mixed | net job gains/losses |
Reading fidelity
high
Study strength
medium
|
n=17
No consensus on net job effects (heterogeneous study results)
|
| The literature shows a surge in research activity on AI and labor markets in 2023–2025 and a concentration of studies in advanced economies. Research Productivity | null_result | publication counts by year and geographic coverage |
Reading fidelity
high
Study strength
medium
|
n=17
Surge in research activity 2023–2025; concentration in advanced economies
|
| There is a widespread consensus across the reviewed literature on the need for worker upskilling, active labor‑market policies, and targeted support for displaced workers. Governance And Regulation | positive | policy recommendations (upskilling / labor-market interventions) |
Reading fidelity
high
Study strength
medium
|
n=17
Widespread consensus on need for upskilling, active labor-market policies, and targeted support
|
| Empirical coverage is limited for low‑income countries; evidence from such settings is scarce. Research Productivity | null_result | geographic representativeness of empirical evidence |
Reading fidelity
high
Study strength
medium
|
n=17
Empirical coverage limited for low-income countries; evidence scarce
|
| Long-term evidence on generative AI’s structural labor‑market effects is scarce; few longitudinal studies exist. Other | null_result | availability of long-term / longitudinal studies on generative AI effects |
Reading fidelity
high
Study strength
medium
|
n=17
|
| Design of this work: a systematic literature review and meta‑synthesis of empirical findings from peer‑reviewed journals (2020–2025), based on 17 publications. Other | null_result | study design / review methodology |
Reading fidelity
high
Study strength
medium
|
n=17
|
| Limitations of the review include the small sample of studies, uneven geographic coverage, heterogeneity in methods across studies, and limited long‑run evidence (especially on generative AI), which complicate causal aggregation. Other | null_result | limitations to causal inference and generalizability |
Reading fidelity
high
Study strength
medium
|
n=17
|
| Policy recommendation: invest in targeted upskilling and reskilling, strengthen active labor‑market policies, and design scalable safety nets to mitigate distributional harms of AI. Governance And Regulation | positive | policy interventions aimed at worker outcomes and distributional effects |
Reading fidelity
medium
Study strength
medium
|
not reported
|
| Broader conclusion: AI has the potential to raise productivity and create value, but without proactive policy the benefits risk being concentrated among skilled workers and firms, exacerbating inequality and regional disparities. Inequality | mixed | productivity gains and distributional outcomes (inequality, regional disparities) |
Reading fidelity
medium
Study strength
medium
|
n=17
|